LOST in America: Evidence on local sales taxes from national panel data

LOST in America: Evidence on local sales taxes from national panel data

Regional Science and Urban Economics 49 (2014) 147–163 Contents lists available at ScienceDirect Regional Science and Urban Economics journal homepa...

5MB Sizes 0 Downloads 25 Views

Regional Science and Urban Economics 49 (2014) 147–163

Contents lists available at ScienceDirect

Regional Science and Urban Economics journal homepage: www.elsevier.com/locate/regec

LOST in America: Evidence on local sales taxes from national panel data☆ David R. Agrawal University of Georgia Department of Economics, 527 Brooks Hall, Athens, GA 30602, United States

a r t i c l e

i n f o

Article history: Received 14 May 2014 Received in revised form 19 September 2014 Accepted 24 September 2014 Available online 2 October 2014 JEL classification: H20 H71 H77 K34 L81 R10 R50

a b s t r a c t This paper studies comprehensive national panel data of local option sales taxes at the monthly frequency. I calculate state-by-month population weighted averages and standard deviations of local sales tax rates. I document ten stylized facts concerning the time series patterns and spatial dynamics of local sales tax rates. The paper then proposes a “tax system” approach to tax competition where states compete on a variety of margins – including restrictions on localities' tax setting authority – that are often ignored by the standard focus on tax rates. Using spatial panel data techniques and the state-by-month population weighted averages, I find a significant association between one state's tax system and its neighboring states' tax systems. © 2014 Elsevier B.V. All rights reserved.

Keywords: Commodity taxation Local public finance Fiscal federalism Spatial tax competition Tax systems Municipal autonomy

1. Introduction It can easily be argued that one of the most significant changes in local public finance (in the United States) over the past half century has been the diminishing role played by local property taxes and the increased importance of other revenue sources. No single revenue mechanism has taken up more of this slack than local option sales taxes (LOST). Local option sales taxes are the second largest own-source of revenue for local governments. On average, LOST raise 12% of municipal revenue (Sjoquist and Stoycheva, 2012; Mikesell, 2010). Yet despite this trend, we know very little about changes in this tax from a national perspective. This paper documents both the time and spatial dynamics of location ☆ The author is also an Affiliate Member of CESifo. Sanjukta Das, Nabaneeta Biswas, and John Kim provided excellent research assistance. Thank you to Adam Cole, Michael Eriksen, Andrew Hayashi, James Hines, William Hoyt, William Lastrapes, Olga Malkova, David Mustard, Edgar Olsen, Gregory Trandel, John Turner, Gary Wagner and David Wildasin for helpful discussions on the topic. Special thanks to the editor and to two anonymous referees who greatly improved the paper. Conference participants at the National Tax Association Meetings and the Southern Economic Association Meetings helped to improve the paper. Thanks to Pro Sales Tax for providing me access to the sales tax data. Any remaining errors are my own. E-mail address: [email protected].

http://dx.doi.org/10.1016/j.regsciurbeco.2014.09.006 0166-0462/© 2014 Elsevier B.V. All rights reserved.

option sales taxes using high frequency national panel data. Both correlations over time and space are necessary to properly understand LOST. Because of the difficulty of assembling a national panel data set of municipal tax rates, many tax incidence and cross-border shopping studies focus on a sub-set of metropolitan areas or a single state. If the researcher only observes state tax rates, the researcher will incorrectly measure tax differentials across time and space. This paper represents the first attempt to assemble comprehensive national panel data on LOST. I have constructed a database of every district, municipal, county, and state sales tax rate in the United States from 2003 to 2011 at the monthly frequency. Armed with this data in hand, I am able to document important yet previously unknown facts concerning these important taxes. The goal of this paper is four-fold. First, I will describe the institutional features of LOST on a state-by-state basis. When working with national data it is important to understand the institutional limitations. In some circumstances, comparisons across states are not valid. As such, I have carefully analyzed state statutes governing LOST and I outline noteworthy institutional structures governing these taxes that vary across states. Second, I develop state-by-month local sales tax rate indexes. These indexes are designed to allow researchers who have state level data to include a measure of the population weighted local

148

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

sales tax rate on top of the statutory state tax rate observed. Third, I use the indexes to document ten stylized facts regarding LOST. How have LOST changed over time? What type of jurisdictions change tax rates? How does the variation of taxes change within states? Has the Great Recession changed tax setting behavior? What are the dynamics of tax rate changes? Despite the simplicity of these questions, we know very little of their answers at the national level. The answers to these questions suggest fruitful future research topics concerning LOST. Finally, the paper turns from documenting time series patterns of LOST to spatial correlations of these changes over time. The empirical application demonstrates that the tax rate series that I construct can be applied to empirically study outcomes across states. The application I study is fiscal competition. Not accounting for local tax data when studying state sales tax competition will inaccurately measure the intensity of spatial relationships across jurisdictions. I also propose a new theory where states compete for mobile shoppers by selecting a tax system as a whole (inclusive of the local tax rates) rather than a single state tax rate. Although municipalities have freedom to choose their tax rates, they do so in an environment that is constrained by the state government. Through these legal constraints, states can help determine whether or not the average municipal tax rate in a state will be high or low — or in states that prohibit LOST, that it will be zero. The spatial pattern in the total local tax rate across states is significant. This suggests that a tax system approach (Slemrod and Gillitzer, 2013; Slemrod and Gillitzer, 2014) is important. These spatial statistics complement the stylized facts documented using the time series. Single state studies of LOST are extremely useful.1 However, without national data, (1) the researcher cannot accurately measure tax discontinuities that vary at borders, (2) cross-state comparisons cannot be made, (3) comparisons exploiting relationships between state institutions and municipal tax rates are much more difficult, and (4) elasticities estimated using state-level data may be biased. How high local tax rates will be and the degree of variation of tax rates within a state will influence behavioral responses and must be accounted for. How flexible the LOST system is (in addition to the state statutory tax rate) influences the extent of cross-border shopping, firm location decisions, employment, tax incidence and other behaviors sensitive to tax policies. In such a context, the results in this paper are useful for thinking about what biases may exist when estimating elasticities without local data and whether single state studies are generalizable. In addition to improving the measurement of tax differentials across states and within states, the data will also be useful to incorporate sales tax components into price indexes essential to measuring quality of life across cities. The Council for Community and Economic Research and its predecessor, the ACCRA, publish consumer price indexes for various urban areas across the United States. However, it is often ignored that these price indexes represent the net price to sellers and not the net price to buyers. In their description of the price indexes, the Council for Community and Economic Research notes that it attempts to “produce an index which adequately measures differences in goods and services costs, rather than to produce an inaccurate measure which attempts to incorporate taxes.” Given that state and local taxes vary substantially across states, not correcting for state and local sales taxes could result in measurement error in these price indexes. While correcting for state sales tax rates may be possible, the data to correct for local sales tax rates is not readily available. For example, if a researcher were correcting the price index for New York based solely on the state tax rate (4%), the researcher would ignore that local tax rates (average: 4.5%) in New York are on average higher than the state tax rate. Carrillo et al. (2014) is an example of a study that produces

1 They are able to isolate particular institutional features of state tax law. In addition, researchers using one particular state have been able to reach back further in time to obtain data on LOST rates and the various outcomes they are interested in, but a shortcoming is that they are not nationally representative.

the best price measures to date, but would benefit from having access to panel data on local sales tax rates.2 The results in this paper will be of use to urban and regional economists seeking to correct price indexes and to public finance economists seeking to more accurately measure behaviors. Further, this paper aims to be a reference piece on LOST. Up until recently, comprehensive studies across municipalities were limited by data availability. The increased access to “big data” at the state and local level allows the researcher to conduct cross-municipal and cross-county studies that exploit a great deal of variation in an open economy setting. 2. Studies of LOST and state sales tax rates In this section I review studies of local option sales taxes with an emphasis on papers that look at tax rates rather than the revenue implications of LOST. For survey pieces please see Fox (2012) and Sjoquist and Stoycheva (2012). Following theoretical work on commodity tax competition (Mintz and Tulkens, 1986; Kanbur and Keen, 1993; Trandel, 1994; Haufler, 1996; Nielsen, 2001)3 several studies analyze tax competition in the presence of local option sales taxes. Most of these studies focus on one particular state.4 Some examples include Zhoa (2005) and Sjoquist et al. (2007) who study tax competition in the state of Georgia. Luna (2003) and Luna et al. (2007) study the rate setting behavior, including the phenomenon of “maxing out,” in the state of Tennessee. Rogers (2004), Burge and Rogers (2011), and Burge and Piper (2012) study fiscal interdependence and local adoption of sales taxes in the state of Oklahoma. Several recent studies exploit national data on LOST. Agrawal (forthcoming), Agrawal (2013a), and Agrawal (2014) use a national cross-section of LOST rates to estimate fiscal reaction functions for both horizontal, diagonal, vertical strategic interactions and interactions with e-commerce. A much broader literature on tax evasion, tax incidence, firm location decisions, and consumer behavioral responses to commodity taxation has exploited variation at the state or metropolitan level. Some of these studies do not have data on local sales tax rates. Others have used selected samples of MSAs or particular border-pairs. Studies exploiting tax differentials at state borders to identify employment effects resulting from sales tax differentials have emphasized state tax rates.5 Many of the previous studies have transformed the public finance literature concerning the effect of sales taxation. Having access to national panel data provides researchers the ability to broaden their samples beyond one particular border and has the potential to allow the researcher to estimate more precise estimates than they would in a world where only the state tax rate is observed. Even if the researcher does not observe all local tax rates in the country, aggregated measures of local tax rates could effectively modify these studies by allowing for a more accurate measure of the average differentials and incentives. 2 Other studies using the ACCRA include Baum-Snow and Pavan, (2012), Dumond et al. (1999), and Winters (2009). Albouy (2012) adjusts the cost of living for state sales tax differences but not for local tax differences. 3 Other studies such as Hoyt (2001) consider the optimal tax considerations of sales taxation in a federation. 4 Other studies such as Benjamin and Dougan (1997), Devereux et al. (2007) and Jacobs et al. (2010) have analyzed commodity tax competition at the state level — although many of these studies focus on excise taxes. 5 As examples, Poterba (1996) studies a sample of fourteen cities and Besley and Rosen (1999) use a sample of approximately 150 cities in the United States; both studies find that the after-tax price increases by the amount of the tax. For examples of border discontinuity designs, see Thompson and Rohlin (2012) and Rohlin et al. (2014). Studies estimating the behavioral response (cross-border shopping) to sales taxes include Mikesell (1970), Fox (1986), Walsh and Jones (1988), and Tosun and Skidmore (2007). Tax evasion has also been studied in the context of the Internet in Goolsbee (2000), Ballard and Lee (2007), and Einav et al. (2014); all of these studies observe some data on either city or county sales tax rates. Cole (2009) uses state tax rates to study the impact of sales tax holidays on both prices and quantities. The volatility of the sales tax has also been studied in the context of a single state by Hou and Seligman (2008) and in the national context with state tax rate changes (Seegert, 2012).

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

3. State institutional details This section documents several important institutional features of LOST; they help guide the researcher on when cross-state comparisons are valid. The implementation of LOST varies extensively across states. As noted in Slemrod and Gillitzer (2013), states design tax systems, not just tax rates — and each of the points below is a part of that tax system. All of these factors may matter to cross-border shoppers — not just the state tax rate. Beyond differences in statutory state tax rates and whether or not local tax rates are allowed, differences across states include: • The levels of government that are given taxing authority by the state may include town, county, and district governments. States may allow some or all of these layers to levy taxes. In the tables to follow, the reader can observe what levels of government are authorized to set tax rates in each state. • What is included in the taxable base varies; this is especially the case with food. Some states exempt food while others do not. In some states, food purchases are exempt from the state tax rate or taxed at a preferential state rate, but the local tax rate still applies. Some states tax prepared food at a higher rate. Consider groceries: groceries are exempt from state and local taxes in Florida, in Georgia food sales are subject to local taxes, and in Tennessee food is subject to a reduced (but positive) state rate plus local taxes. • Beyond food considerations, the LOST tax base may or may not be the same as the state tax base. In states such as Louisiana, parishes have a say over what is subject to LOST; this adds substantial complexity into the tax system. • The procedures concerning how local jurisdiction may pass tax rates differ; these differences may make it harder for localities to adopt LOST in some states. • Some states cap localities by dictating a maximum LOST rate. When the maximum rate is low, many jurisdictions may max out (see Luna et al. (2007) for the case of Tennessee). In other states, localities who want to adopt a LOST must pass a minimum tax rate. Related to this point, some states require all counties to adopt a uniform local option tax rate imposed by the state (example: the Bradley-Burns uniform local tax law in California), but for which the revenue is distributed locally. The state of Virginia geographically differentiates its tax by setting a higher rate in Northern Virginia and Hampton Roads. • In some states, localities have the authority to set any rate they want while in other states, they must pass tax rates in certain increments. As an example, in Ohio counties must pass local tax rates in 1/4 percentage point increments. • In at least one state, the LOST rates may sunset after a certain number of years. In Iowa, the ballot proposal for a LOST may include a sunset clause. • The county tax rates need not be uniform within a county. Whether county tax rates apply in incorporated places, unincorporated places, or both and how tax revenues are distributed accordingly depend on state legislation. For example, in Iowa, the election concerning a local sales tax is countywide, but “the tax only applies in the incorporated areas and the unincorporated area of the county where a majority vote in favor of the local option tax.” • The size of special taxing districts varies. In some states special districts are sub-municipal, but they may also be super municipal — perhaps even corresponding to county borders. Special districts include ambulance or fire districts or transit districts that have special taxing authority. Districts do not mean school districts. • Who is in charge of collecting and administering local rates may be either the state or local governments. For example, in Idaho, cities “may contract with the state tax commission for the collection and administration” or they may use municipal “authority to administer and collect” local sales taxes.

149

• Rules governing use taxes – the tax due when the resident of the assessing town purchases an item that is not subject to his home town's sales tax – also vary. Whether localities are allowed to assess use taxes and how use taxes are enforced vary greatly.6 • Given recent efforts by states to broaden the definition of nexus, the rules of nexus also vary across states. A firm has nexus (generally) if it has a physical presence in the state from which in profits. In states with click-through nexus laws – like Georgia – firms such as Amazon may be required to collect state and local taxes despite not having a conventional physical presence in the state. • Some states have entered into the Streamlined Sales and Use Tax Agreement. This agreement mainly influences tax administration and tax bases, but in its current form does not require states to give up local sales tax rates. Given all of these complexities, one may wonder whether national data is useful to the researcher. Although the econometrician cannot account for all of these varying policies – though state fixed effects help – the researcher can benefit from within state local variation, more precise measurement of tax differentials at state borders, improved measures of taxes within the state, and across-state aggregate variation in the statutory local tax rates. 4. Constructing state data Data on state sales tax rates are readily available and are often used in studies that exploit panel variation and in studies that use state borders as a means of identification. Agrawal (forthcoming) shows a dramatic “level-effect” of state tax rates on local tax rates: on average local sales taxes are 1.2 percentage points lower on the high-tax side of state borders. Although local sales taxes exhibit spatial gradients away from state borders, the overall level effect persists in state based averages. This suggests that understanding the average level of local tax rates across states is important. In this section, I propose a metric to create a weighted average of local sales tax rates and I construct a panel of state level tax aggregates at the monthly frequency.7 To construct state level aggregates of local sales taxes, I merge (following the procedure in the Appendix) data on local sales tax rates from my database with population data from the American Community Survey of the United States Census. The data on local sales tax rates are from ProSales Tax's national database for months from September 2003 to December 2011.8 Although the tax data tells me information about sales tax rates, I do not observe differences in the sales tax base. The data also contains information on use tax rates, which I do not discuss in this paper.9 I do not observe property tax rates or other tax rates. 6 For a discussion of how use tax enforcement may affect optimal tax policy, see Agrawal (2013b). 7 The Tax Foundation website provides, for at least the past several years, state tax rates along with an average local sales tax rate for the state. The advantages of the developed measures in this paper include: they are monthly statistics rather than annual, they include measures of dispersion, they distinguish between the level of the federalist hierarchy (town, county or district) which is extremely important in local public finance, and they are constructed following a weighting method that matches tax rates to places in the Census rather than zip code. The procedure below will allow the researcher to aggregate up to any level of government such as the county or state level. 8 ProSales Tax is a proprietary company that collects local sales tax data and then sells the data to firms with nexus in many different states. For information on the national database, please see http://www.prosalestax.com/. Although the data were provided to me by this firm, substantial cleaning needed to be conducted in order to make the data consistent over time due to border changes, etc. The Appendix describes all of the cleaning done to the data. 9 In the United States, commodity taxation falls under the jurisdiction of both sales and use tax rates. When a consumer purchases a good from a firm with a physical presence in a state, the consumer pays the sales tax rate at the point of transaction. The use tax rate is the rate that a consumer should remit to the government if they purchase an item from a firm outside of the jurisdiction that they reside; most states allow consumers to declare a credit against any sales taxes already paid in another state. In the United States, the use tax is notoriously under-enforced, and as a result, taxation is effectively levied on the basis of the origin of the sale.

150

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

The ideal measure would be an effective tax rate for the average resident of a particular state that accounts for tax evasion opportunities — whether they be on the Internet or in neighboring jurisdictions. However, cross-border flows across all 25,000 localities in the United States are not observable to the researcher. This leads to a necessary assumption: that shopping occurs and is taxed in the place of residence. This assumption is similarly equivalent to assuming that the use tax is effectively enforced. As such, the tax rates in the tables to follow should not be interpreted as the population weighted effective tax rate in the state; rather, they should be interpreted as the population weighted statutory tax rate under the destination principle.10 The destination principle, although not actively enforced, is the legal statutory requirement to pay taxes. Two points of discussion are in order. First, the direction of the bias from defining the weighted averages based on the town of residence is unclear. For states where residents engage in cross-border shopping in order to avoid state and local taxes, the metrics I construct will overestimate the average effective tax rates. However, because 30% of the U.S. population does not live in a Census place and the majority of towns are small, many individuals live in places that do not have retail stores within their home town. As a result, these individuals will shop in nearby larger jurisdictions where the tax rate is likely higher than in their home town.11 Intuitively, because these two effects are offsetting, they are likely to only have a small bias on the aggregated metrics I construct. Second, the choice of population weights merits some discussion. In some sense, this choice is dictated by the data that is available at the national level. One might argue that the amount of retail sales may be a better choice of weights compared to population. Retail sales at the local level are difficult to obtain for all states in the United States. Only some states publish taxable sales data and even if states did publish the data, weighting by a variable that is endogenous to the tax rates would not be desirable. Given that several studies such as Burge and Rogers (2011) indicate that urban and suburban areas attract cross-border shoppers, population weights are also consistent with giving more weight to jurisdictions that have a large amount of sales. Letting i index the municipal level of government, k index states and t index the month, aggregated measures of the sales tax rates can be constructed in the following manner. I estimate E(τk,t) as the population weighted average given by: X τi;k;t pi τk;t ¼

i∈Θk

X

pi

;

ð1Þ

i∈Θk

where τi,k,t denotes the municipal tax rate, pi is the population of a town or unincorporated place i within a particular county, and τ− k,t is the weighted tax rate for state k in month t. Θk is the set of towns in state k; it contains one element for each town i.12 Notice that I construct the weighted averages holding population fixed; I can use either 2000 Census populations or 2010 ACS population numbers. The results from both exercises are similar.13 The population weighted district tax rate

10 Studies using the statutory state tax rate suffer from the same problems; the statutory state tax rate is not the effective tax rate across jurisdictions. Tax incidence (Harding et al., 2012) and evasion (Lovenheim, 2008) may also vary within a state. 11 Standard models of tax competition such as (Kanbur and Keen, 1993) show that larger jurisdictions set higher tax rates. Burge and Rogers (2011) highlight the role of urban– rural tax exportation stories and the loss of LOST tax revenues experienced by rural communities who lose out when their residents shop in cities or suburbs. Brülhart et al. (2012), Jofre-Monseny (2013), and Luthi and Schmidheiny (2014) focus on the relationship between tax policy and agglomeration. 12 Technically, note that i is defined to account for the fact that some towns cross county borders (see Appendix for more details) and some towns are not in Census place data. 13 Annual population data did not become available until the release of the five year ACS data and as a result, to update populations, I would need to linearly interpolate population numbers between 2000 and 2010. Such an exercise is entirely possible, but holding the population fixed at either 2000 or 2010 does not seem to make a difference in the rates.

is also defined using a similar equation. Denote the district tax in town i as δi,k,t and the population weighted average of the district tax rates as δ− k,t. For details on the assignment of districts to towns, please see the Appendix. Because taxation occurs at multiple levels of government, I also define population weighted average county tax rates. Defining Ti,k,t as the county tax rate that applies to town i, the state weighted average can be constructed as X T i;k;t pi T k;t ¼

i∈Θk

X

pi

:

ð2Þ

i∈Θk

The weighting occurs at the town level rater than the county level; this is because in some states, county tax rates have different rates depending on whether towns are home rule or not. In most states, where county tax rates are common within a county, weighting at a more disaggregated level will yield the same weighted average as using the county population weights. Let ζk,t denote the state sales tax rate. Then, the population weighted average of the total tax rate in the state, Ωk;t , is given by: X Ωk;t ¼ ζ k;t þ

i∈Θk

 τi;k;t þ δi;k;t þ T i;k;t pi X

pi

;

ð3Þ

i∈Θk

where it is easy to see that the aggregated measure of all local option sales taxes within a state is simply equal to Ωk;t −ζ k;t . I report Ωk;t and each of the disaggregated measures of town, district and county statistics. I addition to statistics on the population weighted aggregates, measures of dispersion of the data within a state provide useful information. Because most studies of LOST are within a state, we do not know what factors – institutional, economic, or political – drive differences in the variation of LOST rates within a state. Simple scatter plots using a regression discontinuity design in Agrawal (forthcoming) suggest that the standard deviation and the range of LOST are higher in low-tax states than in hightax states. The variance of municipal rates within a state is given by   h  i2 Var τ k;t ¼ E τ k;t −E τk;t

ð4Þ

where the expected value E(τk,t) is given by Eq. (1). The sample analog of this equation for each state is estimated as 2 X  pi τ i;k;t −τ k;t   i∈Θk ^ 1 0 1 ð5Þ Var τ k;t ¼ 0 X 1 @X 2A @ A X pi − pi pi i∈Θk i∈Θk i∈Θk

where the second term in the denominator denotes the bias correction term derived in the Appendix. I report the standard deviation instead of the variance. Similar sample analogs can be defined for all other tax rates. The methods in this paper are used to aggregate up to the state level, but the same procedure applies to aggregating up to the county or national level. Tables 1 to 2 show the above statistics across states for December 2003 and December 2011 using 2010 population weights. Tables for the years in between are all reported in the Appendix; unweighted results and weighting by 2000 populations are also in the Appendix. The population weighted total sub-state tax rate in 2011 was 1.48 percentage points and this value includes zeros for observations where localities are prohibited from implementing LOST. This implies a 30% addition to the average state tax rate. The standard deviation is 1.49. There is substantial variation across the states. For example, in 2011, West Virginia and Mississippi had only one town each utilizing local sales taxes; in West

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

151

Table 1 Tax rates and standard deviations for December 2011. State

State rate

Total local

County

Town

District

Total

sd (total local)

sd (county)

sd (town)

sd (district)

Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming US

4 0 6.6 6 7.25 2.9 6.35 0 6 6 4 4 6 6.25 7 6 6.3 6 4 5 6 6.25 6 6.875 7 4.225 0 5.5 4.6 0 7 5.125 4 4.75 5 5.5 4.5 0 6 7 6 4 7 6.25 4.7 6 4 6.5 6 5 4 5.629

3.999 1.155 2.307 2.123 0.864 4.012 0 0 0 0.647 2.827 0.35 0.019 2.007 0 0.796 1.836 0 4.666 0 0 0 0 0.276 0.003 3.223 0 1.049 3.328 0 0 2.163 4.477 2.103 1.257 1.279 3.427 0 0.337 0 1.172 1.673 2.421 1.596 1.968 0.16 1 2.233 0.002 0.425 1.418 1.482

1.603 0.56 0.698 1.216 0.806 0.719 0 0 0 0.647 2.784 0.35 0.013 0.649 0 0.788 1.036 0 2.243 0 0 0 0 0.034 0 1.306 0 0.002 3.289 0 0 0.26 2.228 2.055 0.135 1.263 0.721 0 0.337 0 0.261 0 2.39 0.13 1.071 0 0.693 0.277 0 0.373 1.418 0.803

2.189 0.595 1.609 0.907 0.059 2.502 0 0 0 0 0.043 0 0.006 0.732 0 0.008 0.789 0 1.747 0 0 0 0 0.118 0.003 1.528 0 1.047 0 0 0 1.838 1.996 0 1.122 0 2.706 0 0 0 0.269 1.521 0.031 1.037 0.324 0.16 0.307 1.728 0.002 0 0 0.542

0.206 0 0 0 0 0.792 0 0 0 0 0 0 0 0.625 0 0 0.011 0 0.676 0 0 0 0 0.124 0 0.389 0 0 0.039 0 0 0.064 0.252 0.048 0 0.015 0 0 0 0 0.642 0.152 0 0.429 0.573 0 0 0.228 0 0.053 0 0.136

7.999 1.155 8.907 8.123 8.114 6.912 6.35 0 6 6.647 6.827 4.35 6.019 8.257 7 6.796 8.136 6 8.666 5 6 6.25 6 7.151 7.003 7.448 0 6.549 7.928 0 7 7.288 8.477 6.853 6.257 6.779 7.927 0 6.337 7 7.172 5.673 9.421 7.846 6.668 6.16 5 8.733 6.002 5.425 5.418 7.11

1.538 2.274 0.89 0.956 0.546 1.456 0 0 0 0.47 0.483 0.267 0.12 1.235 0 0.4 0.72 0 0.647 0 0 0 0 0.296 0.027 0.85 0 0.682 0.358 0 0 1.011 0.533 0.166 0.937 0.448 1.373 0 0.689 0 0.863 0.95 0.244 0.663 0.293 0.347 0 0.597 0.04 0.202 0.734 1.491

0.992 1.549 0.198 0.542 0.539 0.616 0 0 0 0.47 0.415 0.267 0.08 0.59 0 0.405 0.36 0 2.131 0 0 0 0 0.069 0 0.48 0 0.033 0.396 0 0 0.469 2.213 0.104 0.229 0.458 0.569 0 0.689 0 0.441 0 0.323 0.222 0.476 0 0.466 0.559 0 0.22 0.734 1.075

1.686 1.685 0.864 0.847 0.183 1.458 0 0 0 0 0.205 0 0.09 0.576 0 0.086 0.676 0 2.305 0 0 0 0 0.232 0.027 1.076 0 0.684 0 0 0 1.245 2.435 0 0.853 0 1.552 0 0 0 0.512 0.849 0.248 0.679 0.472 0.347 0.466 1.098 0.04 0 0 1.117

0.418 0 0 0 0 0.553 0 0 0 0 0 0 0 0.451 0 0 0.046 0 1.361 0 0 0 0 0.126 0 0.537 0 0 0.1 0 0 0.19 0.194 0.149 0 0.119 0 0 0 0 0.636 0.732 0 0.443 0.289 0 0 0.549 0 0.107 0 0.36

The first six numerical columns present the weighted average of the state, total local (county + town + district), county, town, district, and total combined tax rates. The final four columns present the weighted standard deviations of these rates within a state. All population weights are from the 2010 ACS. The total tax rate is the sum of the first two columns.

Virginia, the first local rate appeared in 2011. At the opposite extreme is a state like Louisiana where the local tax rate is 4.67 relative to a state rate of only 4 percentage points. New York is in a similar situation where the local rate is 4.48 relative to a four percentage point state rate. 5. Facts about LOST 5.1. The time series patterns of LOST Figs. 1–3 document the time series properties of tax rates for states with local taxes.14 Fig. 1 shows the gradual increase in local sales tax rates over time. Excluding non-LOST states shows an increase in level terms of the total local sales tax rate given by Ωk;t −ζ k;t from 1.59 to 14 The first two figures of the Appendix show the same weighted averages for all states — including states that do not allow for LOST.

1.82. Of the various types of local tax rates, county tax rates have risen by 10%, city tax rates have risen by 14%, and district tax rates have become more popular relative to their initial values in 2003. This corresponds to 0.10, 0.09, and 0.04 percentage point changes in county, town, and district tax rates when using only states with LOST, respectively. The population weighted state tax rate has risen by 1.4% of its initial value or 0.08 percentage points. Including states that do not allow for LOST, the total local sales tax rate has risen from 1.29 to 1.48, which corresponds to a 14% change over the time period. State rates increased by 0.16 percentage points when using all states, which is 3% of the initial value. Thus, although state tax rates have risen by a slightly larger amount in level terms, the growth rate in municipal taxes has been more rapid. Fact 1. Local sales tax rates have increased over the last decade. The percent change of local sales tax rates – driven mostly by increases in county and town tax rates – has exceeded the percent change in state tax rates.

152

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

Table 2 Tax rates and standard deviations for December 2003. State

State rate

Total local

County

Town

District

Total

sd (total local)

sd (county)

sd (town)

sd (district)

Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming US

4 0 5.6 5.125 7.25 2.9 6 0 5.75 6 4 4 6 6.25 6 5 5.3 6 4 5 5 5 6 6.5 7 4.225 0 5.5 4.25 0 6 5 4.25 4.5 5 6 4.5 0 6 7 5 4 7 6.25 4.75 6 3.5 6.5 6 5 4 5.47

3.501 1.031 2.098 1.974 0.635 3.461 0 0 0 0.597 2.566 0 0.005 1.241 0 1.155 1.521 0 4.389 0 0 0 0 0.107 0.003 2.351 0 0.986 3.136 0 0.003 1.54 4.081 2.548 0.894 1.122 3.081 0 0.217 0 0.508 1.633 2.359 1.504 1.678 0.007 1 1.826 0 0.392 1.311 1.297

1.56 0.493 0.613 1.171 0.632 0.704 0 0 0 0.597 2.566 0 0 0.383 0 0.616 0.915 0 2.149 0 0 0 0 0.001 0 1.056 0 0 3.097 0 0 0.503 2.074 2.5 0.004 0.938 0.51 0 0.217 0 0.192 0 2.312 0.13 0.898 0 0.69 0.361 0 0.341 1.311 0.726

1.941 0.539 1.485 0.798 0.003 2.321 0 0 0 0 0 0 0.005 0.474 0 0 0.607 0 1.851 0 0 0 0 0.106 0.003 0.983 0 0.986 0 0 0.003 1.037 1.839 0 0.89 0 2.571 0 0 0 0.164 1.495 0.047 1.01 0.419 0.007 0.31 1.112 0 0 0 0.468

0 0 0 0.005 0 0.436 0 0 0 0 0 0 0 0.385 0 0.539 0 0 0.389 0 0 0 0 0 0 0.312 0 0 0.039 0 0 0 0.168 0.048 0 0.184 0 0 0 0 0.153 0.138 0 0.365 0.361 0 0 0.353 0 0.052 0 0.104

7.501 1.031 7.698 7.099 7.885 6.361 6 0 5.75 6.597 6.566 4 6.005 7.491 6 6.155 6.821 6 8.389 5 5 5 6 6.607 7.003 6.576 0 6.486 7.386 0 6.003 6.54 8.331 7.048 5.894 7.122 7.581 0 6.217 7 5.508 5.633 9.359 7.754 6.428 6.007 4.5 8.326 6 5.392 5.311 6.77

1.411 2.072 0.82 0.912 0.372 1.406 0 0 0 0.483 0.65 0 0.108 0.967 0 0.756 0.632 0 0.709 0 0 0 0 0.226 0.027 0.831 0 0.665 0.287 0 0.076 0.567 0.548 0.149 0.709 0.429 1.331 0 0.417 0 0.561 0.969 0.241 0.715 0.25 0.069 0 0.441 0 0.228 0.642 1.366

0.919 1.406 0.336 0.513 0.373 0.733 0 0 0 0.483 0.65 0 0 0.363 0 0.483 0.355 0 2.102 0 0 0 0 0.031 0 0.438 0 0 0.306 0 0 0.312 2.087 0 0.033 0.336 0.53 0 0.417 0 0.397 0 0.33 0.221 0.617 0 0.467 0.613 0 0.235 0.642 1.031

1.533 1.509 0.734 0.791 0.039 1.429 0 0 0 0 0 0 0.108 0.494 0 0.006 0.537 0 2.158 0 0 0 0 0.224 0.027 0.806 0 0.665 0 0 0.076 0.652 2.235 0 0.711 0.009 1.488 0 0 0 0.373 0.842 0.25 0.672 0.621 0.069 0.467 0.852 0 0 0 1.013

0 0 0 0.051 0 0.403 0 0 0 0 0 0 0 0.336 0 0.497 0 0 0.936 0 0 0 0 0 0 0.317 0 0 0.1 0 0 0 0.13 0.149 0 0.32 0 0 0 0 0.363 0.684 0 0.424 0.228 0 0 0.514 0 0.105 0 0.277

The first six numerical columns present the weighted average of the state, total local (county + town + district), county, town, district, and total combined tax rates. The final four columns present the weighted standard deviations of these rates within a state. All population weights are from the 2010 ACS. The total tax rate is the sum of the first two columns.

One noteworthy event is the spike in state-level tax rates after the Great Recession, but not local rates. From Fig. 1 it is also noticeable that the growth in LOST has slowed in recent years. In an effort to study how LOST has grown over time, Fig. 2 plots the percent change in the total local tax rate — defined as the difference of the natural logarithm of the tax rate minus the natural logarithm of the lagged tax rate. Immediately noticeable from this figure is that the percent change in LOST is usually positive. Fact 2. Local sales tax rates have been more stable in the months following the Great Recession than they were prior to or during the recession. The average monthly percentage change in total local tax rates in the period prior to the end of the Great Recession was .17% but it fell to .04% in the period following the end of the Great Recession. A simple test of the difference in group means rejects the hypothesis that the percentage changes are the same. Such a result is surprising in light of conceivable increases of revenue demands on municipal

governments following the recessionary period. However, it is consistent with a view where local sales tax rate increases are less politically feasible in periods when income and spending are relatively low. The results suggest that local sales tax changes may be driven by the economic climate or the business cycle. Although taxes have risen over the sample period, one may wonder what has happened to the overall dispersion of tax rates in the economy. Using the population weighted standard deviations as defined in Eq. (5), I calculate a similar statistic at the national level over the course of the complete time series for the data. Fig. 3 shows how the population weighted standard deviation of LOST has changed over time. The dispersion of local tax rates has increased consistently over time. However, the standard deviation of state plus total local tax rates dropped significantly over the period 2004–2007 but rose dramatically after the Great Recession. This seems to suggest that state total tax rates were much more

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

153

Fig. 1. Tax rates over time. These figures show how the weighted averages of tax rates have evolved over time using data at the monthly frequency. The averages – including the state tax rate data – include only states that allow for local sales tax rates at one or more levels of government. Thus, the separate county, town and district averages will include some states that do not allow for each of these respective taxes. The left figure looks at the time series properties of only local rates, while the right figure looks at the state rates and the total (state + local) rates.

harmonized before the Great Recession, but much less harmonized after the Great Recession. Fact 3. The dispersion of local sales tax rates has increased over time. The dispersion in county tax rates has increased by less than city tax rates. Using all states, the standard deviation in the total local tax rate has increased by about 9% from its initial value in 2003. Of the individual components, the standard deviation of city rates increased by 9.8% while the dispersion of county tax rates increased by 4.8%. The increase in the population weighted dispersion of local sales tax rates could be driven by two factors: (1) jurisdictions electing to change their tax rates may already be in the extreme upper tail of the distribution of tax rates or (2) jurisdictions electing to change their tax rates are only slightly above average but that they are relatively large jurisdictions. In the subsequent sections I will explore these two possibilities. 5.2. Changes across states In the next set of figures, I document changes across states. For this section, I use the state level aggregates displayed in Tables 1 to 2. Fig. 4 shows the relationship between local tax rates and state tax rates in 2003 and 2011. Local tax rates are higher in low-tax states than they are in high-tax states. This confirms the regression discontinuity results in Agrawal (forthcoming). A simple univariate regression indicates that for each percentage point increase in the state tax rate, the average local tax rate is 0.64 percentage points lower.15 This correlation has become more intense over time and is economically significant. Fact 4. Local tax rates are higher in low-tax states than local tax rates in high-tax states. Fig. 5 shows a statistically significant relationship between the standard deviation of local sales tax rates and the state tax rate; the right panel shows its correlation with the average local tax rate in the state. Not only do states with low state tax rates set higher tax rates, but they also have a greater degree of dispersion in their local rates. In the highest tax states, the standard deviation of local tax rates is less than 0.5, but in the lowest tax state it is over 1. However, over time, this negative interdependence between the state tax rate and the dispersion of local sales tax rates has become less intense. This suggests that the dispersion has increased more-so in relatively high-tax states. This could be a result of a number of reasons including the presence of statutory maximums on LOST, which are most likely to bind in states 15 All of the results in this section use a simple univariate regression using across state variation to determine correlations in the data. None of the coefficients on these equations should be viewed as causal effects. Thus, the word effect or relationship does not imply that it is causal.

where LOST are already relatively high — and in fact, past studies have found that “maxing out” is common in these states. The right panel shows that the standard deviation of LOST is highest in states where the average LOST rate is highest. Fact 5. The dispersion of local sales tax rates is greatest in states with low-sales tax rates at the state level and in states with high average local tax rates. Fig. 6 studies the change in the standard deviation from 2003 to 2011. Dispersion has weakly increased the most in states with initially high state tax rates. More pronounced, is the second panel of this figure, which shows a statistically significant positive correlation between the change in the standard deviation and the change in the average tax rate. In states where the average tax rate increases the most, the standard deviation also increases more intensely. For each percentage point increase in the average local tax rate in a state, the standard deviation rises by 0.29. Fact 6. The dispersion of local sales tax rates has increased the most in states where local tax rates increased the most. Fig. 7 shows, at the aggregated level, the relationship between the changes (2003 to 2011) of various tax rates with other levels of local government in the federalist hierarchy. The results indicate that in states where county tax rates rose, municipal tax rates fell, although the association is statistically insignificant. In states where municipal tax rates rose, district taxes also fell. No discernible relationship between district and county tax rates exists.

Fig. 2. Percentage change in tax rates over time. This figure shows how the monthly percentage change of local tax rates has evolved over time. The series starts at the end of 2004. The vertical lines mark the start and end of the Great Recession. The years on the horizontal axis denote December of the year.

154

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

Fig. 3. Standard deviations over time. These figures show how the weighted standard deviations of tax rates have evolved over time using data at the monthly frequency. The standard deviations – including the state tax data – include only states that allow for local sales tax rates at one or more levels of government. Thus, the separate county, town and district standard deviations will include some states that do not allow for each of these respective taxes. The left figure looks at the time series properties of only local taxes, while the right figure looks at the combined local rates and the total (state + local) rates.

Fig. 4. Relationship between local tax rates and state tax rates over time. The figure plots the weighted total local tax rate (town plus county plus district) and the state tax rate for 2003 and 2011. States with a zero state tax rate or no local tax rates are excluded. The line of best fit is from a univariate regression of the same variables.

Fact 7. Aggregated to the state level, changes in the weighted municipal tax rates and changes in the weighted district tax rates are negatively related. The association between changes in the weighted town tax rate and weighted county tax rate is negative, but statistically insignificant.

5.3. The high-frequency dynamics of changes We know relatively little concerning the dynamics of tax competition. 16 How frequently do jurisdictions change their tax rates? Are tax rates sticky? When they do change their rates are they more likely to pass a small marginal change or a dramatic change? What types of jurisdictions are more likely to change their rates? Fig. 8 shows the number of county tax rate changes over time and the average size of tax rate changes.17 When constructing this figure I use only county tax rate changes that affect the entire county. What is noticeable is that increases dominate the number of decreases. Fig. 9 shows the number of people and towns exposed to tax rate changes. 16 Dynamic responses to fiscal policy have been the subject of theoretical work, notably (Wildasin, 2011). 17 State tax changes are included in the Appendix.

Over the course of the sample, approximately 950 county tax changes occur of which approximately 35% are decreases; the total number of counties in states allowing for county sales taxes is approximately 1650. Of these changes, some counties change their tax rate more than once. The average tax change was an increase of .19 percentage points. In an average month, 770,000 people are exposed to a county tax decrease while 900,000 people are exposed to a county tax increase. On average, the sizes of the jurisdictions changing their tax rates are relatively large — having an average population of over 120,000 and a median population of approximately 65,000. Furthermore, the final panel shows that a large number of towns are affected by county tax changes. In an average month, 32 towns are exposed to a county tax rate decrease, while 59 towns are exposed to a county tax increase. The numbers are largest when state level institutional reforms make it so that many or all counties within the state can easily change their tax rates. The results have implications for vertical tax competition. Fact 8. Over the period 2003 to 2011, the total number of town tax changes is approximately 25% of the sample of towns in states with LOST. The total number of county tax changes is over 50% of the sample of towns in states with LOST. Some of these changes correspond to counties and towns that change their tax rate more than once over the sample. Approximately 35% of county tax rate changes are decreases and approximately 20% of town tax rate changes are decreases. Figs. 10 and 11 show similar statistics at the town level. In all of these statistics each matched Census place remains one observation. For non-matched Census places, they appear as one average tax rate for the residual population; if this average changes, I count it as one change.18 Over the course of the sample (100 months), there are just under 5000 town tax changes, of which 20% are decreases. There are 18,765 Census places in America that are in states allowing for LOST. The average tax change was 0.39 percentage points. Approximately 165,000 people are exposed to a town tax decrease in the average month, while 590,000 are exposed to a tax increase in the average month. The average size of a jurisdiction that increases its tax rate is approximately 15,000 people with a median of 7300 people in towns. Again these jurisdictions are bigger than the average jurisdiction in America. Fact 9. Although the number of tax rate changes is relatively small, tax rate changes disproportionately occur in large towns and counties, which implies that a large fraction of the United States population gets exposed to changes in LOST in a given year.

18 This average may change because one or more unmatched towns change its rate. Thus, these numbers represent lower bounds on the total number of changes. The results are robust to excluding these observations.

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

155

Fig. 5. Relationship between the standard deviation and state/local tax rates. The left figure plots the weighted standard deviation of the total local tax rate (town plus county plus district) and the state tax rate for 2003 and 2011. States with a zero state tax rate or no local tax rates are excluded. The right figure plots the same standard deviation with respect to the weighted average of the local tax rates. The line of best fit is from a univariate regression of the same variables.

Fig. 12 tries to see if there is a relationship between the level of the tax rate in the period prior to the change and whether or not a jurisdiction changes its tax rate. What is immediately noticeable is that jurisdictions lowering their tax rate have a much more disperse pattern of rates. Prior to a tax change in the county tax rate, the average county rate was 2.07 in places that subsequently lowered their rate and 1.02 in counties that raised their rate. In towns, the average tax rate prior to a decrease was 1.37; the average tax rate prior to an increase was 0.56. Finally, Fig. 13 shows the correlation between the size of tax rate changes and initial tax rates using disaggregated data. The size of the tax rate change is not at all related to the lagged tax rate in the jurisdiction at the town level, but is related at the county level. There is a strong negative association between the size of the change in the town tax rate and the contemporaneous change in the county tax rate that the town is within. Given the strength of this relationship, this suggests that towns and counties may coordinate their policies — this may be a result of tax rates going into effect in both jurisdictions at the end of the fiscal year or

may be a result of coordinated ballot initiatives to pass a local sales tax. What is noticeable is that most towns that raise their tax rate are in counties that lower their tax rate. Among towns that lower their tax rate, the division is less clear, but still pronounced. Towns with large decreases in their tax rates are clustered in the right quadrant, but for smaller tax rate changes, the pattern begins to blur. Fact 10. Looking at the disaggregated data, towns that lower their tax rate are often in counties that contemporaneously increase their tax rate; towns that increase their tax rate are often in counties that contemporaneously decrease their tax rate. The dynamics discussed in this section using national data can be compared to the state-specific studies of Burge and Piper (2012), Sjoquist et al. (2007), and Burge and Rogers (2014). Using Burge and Piper (2012) as an example, in Oklahoma, jurisdictions that can engage in tax exportation have accelerated adoption patterns. Taxation at the county level also inhibits the adoption of municipal LOST — by extension, a result also consistent with municipal taxes being lower on average in high-tax states.

Fig. 6. Patterns of changes in standard deviations of the total local rates from 2003 to 2011. The left figure plots the change in the weighted standard deviation of the total local tax rate (town plus county plus district) from 2003 to 2011 and the state tax rate. States with a zero state tax rate or no local tax rates are excluded. The right figure plots the same change with respect to the change in the local tax rate. The line of best fit is from a univariate regression of the same variables.

156

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

Fig. 7. State-level patterns of changes in the local rates from 2003 to 2011. The first figure plots the change in the weighted average of the town tax rate from 2003 to 2011 and the change in the weighted average of the county tax rates over the same period. The second figure plots the change in the weighted average of the district tax rate from 2003 to 2011 and the change in the weighted average of the town tax rates within the same period. The right figure plots the change in the weighted average of the district tax rate from 2003 to 2011 and the change in the weighted average of the county tax rates within the same period. States that do not allow for both types of taxes are omitted. The line of best fit is from a univariate regression of the same variables.

6. An application to spatial tax competition

6.1. Spatial patterns of the data

In the above sections, I presented evidence on the time series properties of the data. In this section, I present evidence on spatial patterns in the data.

The formulas used to construct measures of the population weighted tax rates at the state level can be applied to construct weighted averages of local taxes at the county level. After doing so, I calculate the change

Fig. 8. Changes in county tax rates over time — number of changes and size of change. The left figure plots the total number of county tax changes observed in each month. The right panel plots the average unweighted county sales tax rate change. Increases are presented on the positive vertical scale and decreases in the tax rate are plotted on the negative vertical scale. Only county tax changes that apply to the entire county are included.

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

157

Fig. 9. Changes in county tax rates over time — number of towns and people exposed. The first figure plots the total number of people exposed to county tax changes in each month (in 1000s of people). The second panel shows the average size of counties changing a tax rate in each month. The third panel plots the number of towns that are exposed to county tax rate changes by month. Increases are presented on the positive vertical scale and decreases in the tax rate are plotted on the negative vertical scale. Only county tax changes that apply to the entire county are included. A few months with extremely large values are omitted.

in the population weighted average local tax rate for every county in the country for 2011 and 2003. After mapping this data, spatial patterns begin to emerge. Fig. 14 shows changes in the total tax rate (inclusive of LOST). The Appendix includes four additional maps that show the spatial arrangement of district, county, town and total LOST changes separately. Fig. 14 shows that changes in the total tax rate vary both across and within states. In some states the variation in the size of changes within a state is quite similar (for example, Virginia). In the case of Virginia, the change was driven by a state reform with little change at the local level. One reason for this is that Virginia is a strong Dillon's Rule state. In Virginia, if local authority is not explicitly granted to sub-state governments, then it is assumed not to exist. Similarly, historical experiences in North Carolina – specifically, many local government going bankrupt in the Great Depression – have shifted power away from localities to the state government. As such, spatial correlation within a state need not be evidence of tax competition; rather they may simply be evidence that the state government has strong control over its local governments. Other states such as Alabama, Louisiana, Missouri, Oklahoma, and Colorado show a high degree of variation within a state. These states allow localities much more autonomy when setting LOST compared to Virginia and North Carolina. Within states, some clustering begins to emerge. For example, large decreases are in northeastern Utah and several contiguous counties in south eastern Colorado. Border counties in Arizona and New Mexico both saw increases. Border counties in Oklahoma and Arkansas saw increases; this border contains tourism driven areas. The same is true of three of the states bordering Minnesota. This ocular evidence seems to suggest that spatial correlation of

tax changes may persist across state borders. One reason why this may be the case is that many large metropolitan areas are located on state borders in the United States. Approximately fifty metropolitan areas containing 75 million people reside in multiple-state urban areas. Issues concerning tax exporting may also be important in these border regions.

6.2. A general model Following the work of Case et al. (1993), several empirical studies, have studied tax competition across states.19 I use the state by month data that I have created to study spatial tax competition across states. In order to accurately measure the strategic interactions across governments of the same level, the researcher must also account for vertical strategic interactions. Vertical strategic interactions occur when different levels of government that cohabit the same tax base have tax setting authority. Omission of the vertical interactions may bias horizontal strategic interactions (Besley and Rosen, 1998). Studies of tax competition at the state level often lack information on tax rates at the local level within a state. Going forward, I propose using the population weighted local tax rates within a state to estimate the horizontal strategic interactions across states. Given that state governments likely pay much more attention to the largest cities (places with the largest tax bases), the population weighted averages constructed in this paper are a natural starting point. 19

For excellent summaries, please see Brueckner (2003) and Revelli (2012).

158

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

Fig. 10. Changes in town tax rates over time — number of changes and size of change. The first figure plots the total number of town tax changes observed in each month. The second panel plots the average county sales tax rate change. Increases are presented on the positive vertical scale and decreases in the tax rate are plotted on the negative vertical scale. The third panel shows the average tax change where increases and decreases are not calculated separately.

Relatively few studies of tax rates employ spatial and high-frequency temporal data to identify tax competition. Thus, before proceeding, I summarize a general form of a spatial econometric model using panel data; in the following sections, I discuss the restrictions I impose. Consider a world where I let i index a state, let j index a neighbor, t index time, and let c index C possible control variables denoted x. A general specification for panel data can be written as τi;t

C C X X X X ¼ α þ φτi;t−1 þ ρ wi j τ j;t þ xi;t;c βc þ wi j x j;t;c μ c þ ζ i c¼1 c¼1 j≠i j≠i X þ γt þ λ wi j v j;t þ εi;t ;

ð6Þ

j≠i

where τi,t is various measures of the tax rate aggregates in state i at time t discussed below. State fixed effects (ζi) and time dummies (γt) are also included. To allow for strategic interactions across space, a spatial lag ∑ j ≠ i wijτj,t is included where wij is the weight that neighboring state j is given. Following the literature, I restrict ∑ j ≠ i wij = 1 for each state i. The term with the double summation allows for weighted averages of your neighbors' control variables (x) to influence your own tax rate. Let v be an error that has a spatial process. If a one period (time) lagged dependent variable is included, the model is dynamic. Exclusion of time lag implies that the model is static. 6.3. Implementing the model First, the choice of weights wij merits some discussion. For purposes of this paper, I will use a measure of contiguity. With uniform contiguity weights, each neighbor is given equal weight. As an example, consider the state of Connecticut, which has three neighbors. The

weights given to New York, Massachusetts, and Rhode Island would be 1/3 each; all other states are given zero weight. For the case of sales tax rates, contiguity seems to be reasonable given that states are worried about cross-border shopping. Unlike the case of cigarettes, smuggling of retail goods is not organized and usually only occurs across neighboring state borders. More formally, I can define the row-normalized uniform contiguity weights as 8 < 1 wi j ¼ mi : 0

if j ∈ M i if j ∉ M i

ð7Þ

where Mi is the set of towns contiguous to state i and where mi is the number of towns in M i . As a robustness check, I alternatively define the weights to give longer contiguous borders more weight than shorter borders where 8 ℓi j > > > : j∈Mi 0

if j ∈ M i

ð8Þ

if j ∉ Mi

and ℓij is the length of the border between states i and j.20 Second, to estimate the model, I impose some restrictions on Eq. (6). I set λ = 0 and estimate what is referred to as a spatial Durbin model (SDM). The spatial Durbin model is advocated by LeSage and Pace (2009) and Elhorst (2010). I choose the SDM in this case because I have relatively few controls that vary at the monthly frequency and because the SDM produces unbiased coefficient estimates if the true 20

The border length data are from Holmes (1998).

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

159

Fig. 11. Changes in town tax rates over time — number of people exposed. The left figure plots the total number of people exposed to town tax changes in each month (in 1000s of people). The right panel shows the average size of towns changing a tax rate in each month. Increases are presented on the positive vertical scale and decreases in the tax rate are plotted on the negative vertical scale. A few months with extremely large values are omitted.

Fig. 12. Changes in tax rates over time — relationship to initial value. The left figure plots the average county tax rate in the period prior to a county changing its tax rate. The right panel plots the average town tax rate in the period prior to a town changing its tax rate. Increases are presented on the positive vertical scale and decreases are plotted on the negative vertical scale.

underlying process is a spatial lag or a spatial error model (Elhorst, 2010). Note further that when the model has φ = 0, it is static. Allowing for a dynamic model would add additional complications. In a dynamic panel data model with fixed individual effects, the autoregressive coefficient is biased (Hsiao, 1986). Further, Lee and Yu (2009) emphasizes the instability concern associated with the situation where the sum of the spatial and dynamic effect estimates exceeds one — which is likely given persistence in the data. In what follows, I estimate Eq. (6) with the noted restrictions using maximum likelihood methods. Following Revelli (2001), I estimate the model in first differences where all variables are differenced. Differencing also has the advantage of removing any linear time trend. State fixed effects are differenced out. Exploiting the high frequency nature of the panel means that relatively few control variables will be tracked at the monthly frequency. As controls, I include data from the Bureau of Labor Statistics (BLS): the monthly unemployment rate of each state and the log of the size of the labor force. In some specifications I also include variables from the American Community Survey (ACS). Unfortunately, the ACS one year estimates start in 2006, so any specifications using ACS data will need to drop several months of data and thus are not strictly comparable to the baseline model. 21 Descriptive statistics of all control variables included in the estimating equation are detailed in the Appendix.

21 Because the model is estimated in first differences, I linearly interpolate the ACS assuming that the data in the ACS is valid for the month of June in the given year.

The results of this model should be interpreted with caution. Identifying tax competition as the causal channel through which spatial patterns arise can be difficult (McMillen, 2010; Gibbons and Overman, 2012). In the most cautionary case, coefficient estimates from the econometric model are simply informative of spatial correlation in the tax rates because unobservable correlated shocks may hit contiguous jurisdictions and endogeneity concerns may be an important issue. Recent developments suggest that, in order to make statements that these spatial correlations are evidence of tax competition, studies must construct a more quasi-experimental design (Agrawal, forthcoming; Eugster and Parchet, 2013; Lyytikäinen, 2012). Thus, the purpose of this empirical model is to show that a researcher with only state data and without access to local data – using a standard reduced form reaction function – may obtained biased estimates of the association between one state's tax rate and its neighboring states' tax rates. Accounting for these local rates will allow the researcher to better measure horizontal competition across states. The estimates help to provide geo-spatial descriptive statistics designed to be complementary to the time series statistics. When using the total rate as the dependent variable, these associations represent spatial relationships of the tax systems across states; when using the state sales tax rate, they represent only the relationship with neighboring state tax rates. At the same time, the high-frequency panel data used in this paper helps. A threat to identification is high-frequency spatially correlated shocks across states. If such shocks are uncommon and the model specified correctly then the equation can be interpreted as a reduced form reaction function and the spatial relationships can be directly interpreted as tax competition.

160

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

Fig. 13. Changes in town and county tax rates — initial values and different levels of government. The first figure plots the county tax rate change on the vertical axis and the tax rate in the county in the period prior to a change. The second figure plots the town tax rate change on the vertical axis and the tax rate in the town in the period prior to a change. The third figure plots the town tax rate change on the vertical axis and the contemporaneous county tax rate change affecting that town. This figure only includes towns that change their own tax rate. The line of best fit is from a univariate regression using population weights.

6.4. Motivation: spatial patterns in the total rate Traditionally, it is assumed that states select their state tax rate in competition with other states. But, what if states do not choose the state tax rate, but rather the total (state plus local tax rate)? In the U.S. Supreme Court ruling Clinton v. Cedar Rapids and the Missouri River Railroad, the Court wrote that “Municipal corporations owe their origin to, and derive their powers and rights wholly from, the legislature. It breathes into them the breath of life, without which they cannot exist. As it creates, so may it destroy. If it may destroy, it may abridge and control.” This principle has become known as Dillon's Rule — that municipalities are creatures of the state.22 I raise the principle underlying Dillon's Rule to simply motivate the premise that although most state governments do not dictate the level of LOST rates, they do control how municipalities and counties are allowed to set local tax rates. As noted in the section governing the institutional differences across states, states may impose restrictions – such as statutory maximums – that may constrain their municipalities. Many of these institutional features will in turn influence the level of the statutory tax rate set by municipalities. In states with restrictive LOST systems, local rates will likely be lower than in states with flexible systems, all else equal. This 22 For legal discussion of Dillon's Rule and the authority of local governments more generally, please see Cohn (1957), Schragger (2001), and Schragger (2006).

result is not because the municipality has a desire to set a competitively low rate, but rather because the state in competition with other states has implicitly dictated that the rate must be low. The weighted average of local plus state tax rates is a (imperfect) gauge of the true effective local tax rate that the state desires within its borders. Although municipalities have freedom to choose their tax rates, they do so in an environment that is dictated by the state government. If states pick a tax system rather than simply a state sales tax rate, then the researcher should study state competition over the components of the local rate rather than the state rate in isolation. One advantage of this procedure is that it allows the researcher to then estimate a spatial equation that does not include vertical strategic interactions. A second advantage is that using the total rate rather than the state rate and the local rate separately allows for the researcher to account for other institutional aspects when estimating spatial associations in the data. In proceeding, I assume that the “tax system” perspective applies to the sales tax system; I cannot say how the sales tax interacts with other local taxes which may also be influenced by state level policy.23

23 Of course, I implicitly assume that all of the strategic reaction occurs through the sales tax system. In reality, states may adjust many different levers: for example they may restrict local sales tax autonomy or property tax autonomy. An even more comprehensive tax system perspective would account for multiple tax rates. The issue of multiple tax rates and how they interact remains a fruitful area of research.

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

161

Fig. 14. Total (state plus local) tax changes, 2003 to 2011. This figure shows changes in the total (state plus each county's population weighted county, town, and district) tax rates from 2003 to 2011. Hashed counties are counties where total tax rates are the same in 2011 as they were in 2003. Hallow counties are counties in states with no sales tax rates (at any level of government). Darker colors imply larger tax increases and lighter colors imply larger tax decreases. (For easier interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

7. Results

setting legal restrictions on the degree of autonomy that the state grants to the localities). In this second view, there is no need to control for the tax rates of other levels of government because the state internalizes them — and as such, the only tax variable on the right hand side of the estimating equation is the spatial lag of the total rate. The coefficient of interest is ρ, which determines the intensity of the spatial associations of tax rates across states. If the coefficient is positive, an increase in your neighbors' tax rate implies that your tax rate will also rise. Looking at column (1a) which uses the spatial weights in Eq. (7), a one percentage point increase in neighbor's state tax rate

Table 3 reports the results for various specifications. I present two sets of comparisons where: (1) the dependent variable is the statutory state sales tax rate and (2) the dependent variable is the total tax rate, which is equal to the state tax rate plus the weighted average of all local tax rates in the state. Use of the statutory sales tax rate is the conventional dependent variable in the literature. When I use the total tax rate in the state, I take the view that states choose their state tax rate as well as the average local tax rate in the state (by effectively Table 3 Strategic interactions from a spatial Durbin model: state rate vs. total rate.

ρ: spatial lag coefficient Number of observations Groups Panel length BLS controls ACS controls Month fixed effects Spatial weights: contiguity (C) or border length (L) Regression weights by (state) population

Without ACS controls: full sample

With ACS controls: sample starts in 2006

State tax rate

State tax rate

Total tax rate

Total tax rate

(1a)

(1b)

(1c)

(2a)

(2b)

(2c)

(3a)

(3b)

(3c)

(4a)

(4b)

(4c)

.057*** (.011) 4802 49 98 Y N Y C N

.058*** (.011) 4802 49 98 Y N Y C Y

.046*** (.009) 4802 49 98 Y N Y L N

.061*** (.013) 4802 49 98 Y N Y C N

.063*** (.013) 4802 49 98 Y N Y C Y

.046*** (.011) 4802 49 98 Y N Y L N

.057*** (.011) 2891 49 59 Y Y Y C N

.057*** (.012) 2891 49 59 Y Y Y C Y

.036*** (.013) 2891 49 59 Y Y Y L N

.051*** (.013) 2891 49 59 Y Y Y C N

.052*** (.014) 2891 49 59 Y Y Y C Y

.040*** (.013) 2891 49 59 Y Y Y L N

In columns (1a–c) and (3a–c) the dependent variable is the state tax rate. In columns (2a–c) and (4a–c) the dependent variable is the total tax rate (all local rates plus state). Columns (a) use uniform contiguity spatial weights and an unweighted regression, columns (b) use uniform contiguity spatial weights and weight the regressions by population, and columns (c) use border length spatial weights in an unweighted regression. Columns (1)–(2) include BLS controls and columns (3)–(4) include ACS and BLS controls. I estimate a spatial Durbin model in first differences; thus, state fixed effects are differenced out. All standard errors are clustered at the state level. ***99%, **95%, *90% .

162

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

Table 4 Strategic interactions from a spatial Durbin model: levels of local taxes. Without ACS controls: full sample Total local rate

ρ: spatial lag coefficient Number of observations Groups Panel length BLS controls ACS controls Month fixed effects Spatial weights: contiguity (C) or border length (L) Regression weights by (state) population

County rate

Town rate

District rate

(1a)

(1b)

(1c)

(2a)

(2b)

(2c)

(3a)

(3b)

(3c)

(4a)

(4b)

(4c)

.152 (.093) 4802 49 98 Y N Y C N

.158 (.097) 4802 49 98 Y N Y C Y

.115* (.067) 4802 49 98 Y N Y L N

.012 (.018) 4802 49 98 Y N Y C N

.012 (.017) 4802 49 98 Y N Y C Y

.022 (.013) 4802 49 98 Y N Y L N

.051*** (.016) 4802 49 98 Y N Y C N

.052*** (.016) 4802 49 98 Y N Y C Y

.034** (.015) 4802 49 98 Y N Y L N

.046* (.024) 4802 49 98 Y N Y C N

.046** (.022) 4802 49 98 Y N Y C Y

.037** (.017) 4802 49 98 Y N Y L N

In column (1) the dependent variable is the weighted average total local tax rate and in columns (2)–(4) it is the local rate listed in the header row of the table. Columns (a) use uniform contiguity spatial weights and an unweighted regression, columns (b) use uniform contiguity spatial weights and weight the regressions by population, and columns (c) use border length spatial weights in an unweighted regression. All columns include BLS controls. I estimate a spatial Durbin model in first differences; thus, state fixed effects are differenced out. All standard errors are clustered at the state level. ***99%, **95%, *90%.

implies a 0.057 percentage point increase in the own state tax rate. In column (1c), I use the border length spatial weights of Eq. (8) and the coefficient falls slightly. Adding ACS demographic and population controls does not change the estimates much. In columns (2a–c), I look for competition in the total tax rate across states. This approach accounts for local sales taxes. The coefficient on the spatial lag is slightly larger than the value of ρ in columns (1a–c). A one percentage point increase in the neighbor's total state plus local tax rate raises the total own tax rate by 0.061 percentage points. The coefficient remains strongly significant which suggests that localities do not counteract the similarities in the first specifications; states are similar to their neighbors both in their state tax rates and in their tax system as a whole. This demonstrates that there are significant spatial relationships even after accounting for local sales taxes, which by fact 4 are lower in high-tax states. Table 4 addresses the channels at work by studying the relationship between the average total local tax rate in a state and that of its contiguous neighbors. Here the coefficient on the spatial lag is 2.5 times larger than the statutory state rate specification. Keep in mind that this specification regresses the average local tax rate in a state on the average local tax rate of the neighboring states; the regression is not identifying spatial relationships within a state. Despite states being large geographically, nearby states are similar with respect to their average local tax rate. The result is especially noteworthy given that local taxes in low-state tax states are higher than local taxes in relatively high-state tax neighboring states. Because states are large and many localities are away from borders, the implication is that the spatial pattern is not necessarily picking up stronger tax mimicking among border jurisdictions on either side of state boundaries. Furthermore, I can break down the relationships across various types of local taxes. Table 4 shows that the relationship is more intense for some levels of government than others; however, the sum of the spatial lag coefficients across county, towns, and districts is smaller than the spatial lag regression using the sum of the tax rates. This suggests that it is the local tax system as a whole that exhibits a stronger spatial association with neighboring states than any one particular component of the local tax system. One explanation is that states elect to set similar institutional rules governing LOST (on average). Consider the simple decision of whether to allow for LOST or not: states not allowing for LOST are concentrated in particular geographic regions. This provides powerful evidence of spatial correlation in the tax system as a whole. This section has shown that aggregate indexes of LOST rates are important to estimating the strategic interactions of states. At a minimum, they act as an effective way of accounting for taxation at multiple levels of government (as in Table 3 columns 2a–c) by allowing the researcher to more accurately measure the horizontal spatial relationships in the

state tax rates. At their best, they allow the researcher to make broader conclusions about how states interact strategically from a tax system perspective where local governments are creatures of the state to varying degrees.

8. Conclusion A national panel of local sales tax rates has eluded researchers — despite its importance to answering fundamental public and urban economics questions. This paper documents what we can learn from such data. Beyond tax competition, use of the indexes constructed in this paper will prove fruitful to study questions concerning firm location decisions, behavioral responses to taxes, tax evasion, and tax incidence. Consider tax incidence as an example. In a general sense, we regress pre-tax prices on a variety of controls and the sales tax rate. We would like that sales tax rate to include local taxes, but until now, that has not really been possible beyond using sub-samples of urban areas or states. If we omit local taxes and use only the state sales tax rate, the coefficient estimate on the sales tax rate will then be missing the covariance between the local rates and pre-tax prices, the variance of local rates, and the covariance of state and local rates. The facts documenting the relationship between local sales taxes and state taxes along with my own previous work imply that the covariance between state rates and local rates is negative. In future work, I hope authors make progress to shed light on the covariance between local rates and pre-tax prices; in doing so, we can reduce the bias in our estimates of tax incidence. Other extensions abound. This study documents ten stylized facts that provide researchers with ample new puzzles and questions. Future research may consider question such as the following. Why have taxes increased faster at some levels of government to others? How can we think about tax systems in the context of tax competition? How do the institutional features across states affect tax competition? Does accounting for local sales tax rates change the estimates of the behavioral responses or tax incidence that we have estimated in the past using only state tax rates or a sub-sample of towns? These and many more questions arise from a first look at national panel data on LOST. In documenting the facts, this study has in some sense raised more questions than answers — and has hopefully generated interest in using “big data” at the municipal level to answer these central public and urban economics questions.

Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.regsciurbeco.2014.09.006.

D.R. Agrawal / Regional Science and Urban Economics 49 (2014) 147–163

References Agrawal, David R., 2013a. Inter-federation competition: sales taxation with multiple federations. Working Paper. University of Georgia. Agrawal, David R., 2013b. Over the borderline: how the characteristics of lines shape optimal tax policy. Econ. Lett. 119 (2), 113–116. Agrawal, David R., 2014. The Internet as a tax haven?: the effect of the Internet on tax competition. University of Georgia Working Paper. Agrawal, David R., 2014. The tax gradient: spatial aspects of fiscal competition. Am. Econ. J. Econ. Pol. (forthcoming). Albouy, David Y., 2012. Are big cities bad places to live? Estimating quality of life across metropolitan areas. Working Paper. Ballard, Charles L., Lee, Jaimin, 2007. Internet purchases, cross-border shopping, and sales taxes. Natl. Tax J. 55 (4), 711–725. Baum-Snow, Nathaniel, Pavan, Ronni, 2012. Understanding the city size wage gap. Rev. Econ. Stud. 79 (1), 88–127. Benjamin, Daniel K., Dougan, William R., 1997. Efficient excise taxation: the evidence from cigarettes. J. Law Econ. 40 (1), 113–136. Besley, Timothy J., Rosen, Harvey S., 1998. Vertical externalities in tax setting: evidence from gasoline and cigarettes. J. Public Econ. 70, 383–398. Besley, Timothy J., Rosen, Harvey S., 1999. Sales taxes and prices: an empirical analysis. Natl. Tax J. 52 (2), 157–178. Brueckner, Jan K., 2003. Strategic interaction among governments: an overview of empirical studies. Int. Reg. Sci. Rev. 26, 175–188. Brülhart, Marius, Jametti, Mario, Schmidheiny, Kurt, 2012. Do agglomeration economies reduce the sensitivity of firm location to tax differentials? Econ. J. 122 (563), 1069–1093. Burge, Gregory S., Piper, Brian, 2012. Strategic fiscal interdependence: county and municipal adoptions of local option sales taxes. Natl. Tax J. 65 (2), 387–416. Burge, Gregory, Rogers, Cynthia, 2011. Local option sales taxes and consumer spending patterns: fiscal interdependence under multi-tiered local taxation. Reg. Sci. Urban Econ. 41, 46–58. Burge Gregory, S., Rogers Cynthia, L., 2014. Identifying Leaders and Followers in Models of Local Tax Competition. University of Oklahoma Working Paper. Carrillo, Paul E., Early, Dirk W., Olsen Edgar, O., 2014. A panel of price indices for housing services, other goods and all goods for all areas in the United States, 1982–2010. J. Hous. Econ. 26, 81–93. Case, Anne C., Hines, James R., Rosen, Harvey S., 1993. Budget spillovers and fiscal policy interdependence. J. Public Econ. 52 (3), 285–307. Cohn, Rubin C., 1957. Municipal revenue powers in the context of constitutional home rule. Northwest. Law Rev. 51 (1), 27–42. Cole, Adam J., 2009. Christmas in August: prices and quantities during sales tax holidays. University of Michigan Working Paper. Devereux, Michael P., Lockwood, Ben, Redoano, Michela, 2007. Horizontal and vertical indirect tax competition: theory and some evidence from the USA. J. Public Econ. 91, 451–479. Dumond, J. Michael, Hirsch, Barry T., David A., Macpherson, 1999. Wage differentials across labor markets and workers: does cost of living matter? Econ. Inq. 37 (4), 4. Einav, Liran, Knoepfle, Dan, Levin, Jonathan, Sundaresan, Neel, 2014. Sales taxes and Internet commerce. Am. Econ. Rev. 104 (1), 1–26. Elhorst, J. Paul, 2010. Applied spatial econometrics: raising the bar. Spat. Econ. Anal. 5 (1), 9–28. Eugster, Beatrix, Parchet, Raphäel, 2013. Culture and taxes: toward identifying tax competition. Discussion Paper No. 2013-39. Fox, William F., 1986. Tax structure and the location of economic activity along state borders. Natl. Tax J. 39 (4), 387–401. Fox, William F., 2012. Retail sales and use taxation. In: Ebel, Robert D., Petersen, John E. (Eds.), The Oxford Handbook of State and Local Government Finance. Oxford University Press, pp. 406–428. Gibbons, Stephen, Overman, Henry G., 2012. Mostly pointless spatial econometrics? J. Reg. Sci. 52, 172–191. Goolsbee, Austan D., 2000. In a world without borders: the impact of taxes on Internet commerce. Q. J. Econ. 115 (2), 561–576. Harding, Matthew, Leibtag, Ephraim, Lovenheim, Michael, 2012. The heterogeneous geographic and socioeconomic incidence of cigarette taxes: evidence from Nielsen Homescan Data. Am. Econ. J. Econ. Pol. 4 (4), 169–198. Haufler, Andreas, 1996. Tax coordination with different preferences for public goods: conflict or harmony of interest. Int. Tax Public Financ. 3 (1), 5–28. Holmes, Thomas J., 1998. The effect of state policy on the location of manufacturing: evidence from state borders. J. Polit. Econ. 106 (4), 667–705. Hou, Yilin, Seligman, Jason, 2008. LOST stability? Consumption taxes and the cyclical variability of state and local revenues. Proceedings of the 100th Annual Meeting of the National Tax Association.

163

Hoyt, William, 2001. Tax policy coordination, vertical externalities, and optimal taxation in a system of hierarchical governments. J. Urban Econ. 50, 491–516. Hsiao, Cheng, 1986. Analysis of Panel Data. Cambridge University Press. Jacobs, Jan P.A.M., Ligthart, Jenny, Vrijburg, Hendrik, 2010. Consumption tax competition among governments: evidence from the United States. Int. Tax Public Financ. 17, 271–294. Jofre-Monseny, Jordi, 2013. Is agglomeration taxable? J. Econ. Geogr. 13 (1), 177–201. Kanbur, Ravi, Keen, Michael, 1993. Jeux sans frontières: tax competition and tax coordination when countries differ in size. Am. Econ. Rev. 83 (4), 877–892. Lee, Lung-fei, Yu, Jihai, 2009. Some recent developments in spatial panel data models. Reg. Sci. Urban Econ. 40 (5), 255–271. LeSage, James, Pace, Robert Kelley, 2009. Introduction to Spatial Econometrics. Taylor & Francis, Boca Raton. Lovenheim, Michael F., 2008. How far to the border?: The extent and impact of crossborder casual cigarette smuggling. Natl. Tax J. 61 (1), 7–33. Luna, LeAnn, 2003. Local sales tax competition and the effect on county governments' tax rates and tax bases. J. Am. Tax. Assoc. 26 (1), 43–67. Luna, LeAnn, Bruce, Donald J., Hawkins, Richard, 2007. Maxing out: an analysis of local option sales tax rate increases. Natl. Tax J. LX (1), 45–63. Luthi, Eva, Schmidheiny, Kurt, 2014. The effect of agglomeration size on local taxes. J. Econ. Geogr. 14 (2), 265–287. Lyytikäinen, Teemu, 2012. Tax competition among local governments: evidence from a property tax reform in Finland. J. Public Econ. 96 (7–8), 584–595. McMillen, Daniel, 2010. Issues in spatial data analysis. J. Reg. Sci. 50, 119–141. Mikesell, John L., 1970. Central cities and sales tax rate differentials: the border city problem. Natl. Tax J. 23 (2), 206–213. Mikesell, John L., 2010. The contribution of local sales and income taxes to fiscal autonomy. In: Ingram, Gregory K., Hong, Yu-Hung (Eds.), Municipal Revenues and Land Policies. Lincoln Institute of Land Policy. Mintz, Jack, Tulkens, Henry, 1986. Commodity tax competition between member states of a federation: equilibrium and efficiency. J. Public Econ. 29 (2), 132–172. Nielsen, Søren Bo, 2001. A simple model of commodity taxation and cross-border shopping. Scand. J. Econ. 103 (4), 599–623. Poterba, James M., 1996. Retail price reactions to changes in state and local sales taxes. Natl. Tax J. 49 (2), 165–176. Revelli, Frederico, 2001. Spatial patterns in local taxation: tax mimicking or error mimicking? Appl. Econ. 33 (9), 1101–1107. Revelli, Federico, 2012. Geografiscal federalism. Working Paper in Preparation for the Handbook of Fiscal Federalism. Rogers, Cynthia L., 2004. Local option sales tax (LOST) policy on the urban fringe. J. Reg. Anal. Pol. 34 (1), 28–50. Rohlin, Shawn, Rosenthal, Stuart S., Ross, Amanda, 2014. Tax avoidance and business location in a state border model. J. Urban Econ. 83, 34–49. Schragger, Richard C., 2001. The limits of localism. Mich. Law Rev. 100 (2), 371–472. Schragger, Richard C., 2006. Can strong mayors empower weak cities? on the power of local executives in a federal system. Yale Law J. 115 (9), 2542–2578. Seegert, Nathan, 2012. Optimal taxation with volatility: a theoretical and empirical decomposition. Working Paper. Sjoquist, David D., Stoycheva, Rayna, 2012. Local revenue diversification: user charges, sales taxes, and income taxes. In: Ebel, Robert D., Petersen, John E. (Eds.), The Oxford Handbook of State and Local Government Finance. Oxford University Press, pp. 429–462. Sjoquist, David L., Smith, William, Wallace, Sally, Walker, Mary Beth, 2007. An analysis of the time to adoption of local sales taxes: a duration model approach. Public Budg. Financ. 27 (1), 20–40. Slemrod, Joel, Gillitzer, Christian, 2013. Insights from a tax-system perspective. CESifo Econ. Stud. 60 (1), 1–31. Slemrod, Joel, Gillitzer, Christian, 2014. Tax Systems. MIT Press. Thompson, Jeffrey P., Rohlin, Shawn, 2012. The effect of sales taxes on employment: new evidence from cross-border panel data analysis. Natl. Tax J. 65 (4), 1023–1042. Tosun, Mehmet S., Skidmore, Mark L., 2007. Cross-border shopping and the sales tax: an examination of food purchases in West Virginia. B.E. J. Econ. Anal. Pol. 7 (1), 1–18. Trandel, Gregory, 1994. Interstate commodity tax differentials and the distribution of residents. J. Public Econ. 53 (3), 435–457. Walsh, Michael, Jones, Jonathan, 1988. More evidence on the border tax effect: the case of West Virginia, 1979–84. Natl. Tax J. 41 (2), 261–265. Wildasin, David, 2011. Fiscal competition for imperfectly-mobile labor and capital: a comparative dynamic analysis. J. Public Econ. 95, 1312–1321. Winters, John V., 2009. Wages and prices: are workers fully compensated for cost of living differences? Reg. Sci. Urban Econ. 39 (5), 632–643. Zhoa, Zhirong, 2005. Motivations, obstacles, and resources: the adoption of the general purpose local option sales tax in Georgia counties. Public Financ. Rev. 33 (6), 721–746.